Constraints on mode couplings and modulation of the CMB with WMAP data.

Simon Prunet∗
Institut d’Astrophysique de Paris, 98 bis, Bd Arago, 75014 Paris, France.

Jean-Philippe Uzan†
Laboratoire de Physique Théorique, CNRS–UMR 8627,
Bât. 210, Université Paris XI,
F-91405 Orsay Cedex, France,
and
Institut d’Astrophysique de Paris,
GReCO, CNRS-FRE 2435,
98 bis Bd Arago, 75014 Paris, France.
arXiv:astro-ph/0406364v1 16 Jun 2004

Francis Bernardeau‡ and Tristan Brunier§

Service de Physique Théorique, CEA/DSM/SPhT,
Unité de recherche associée au CNRS, CEA/Saclay 91191 Gif-sur-Yvette cédex
(Dated: 5 juin 2004)
We investigate a possible asymmetry in the statistical properties of the cosmic microwave back-
ground temperature field and to do so we construct an estimator aiming at detecting a dipolar
modulation. Such a modulation is found to induce correlations between multipoles with ∆ℓ = 1.
Applying this estimator, to the V and W bands of the WMAP data, we found a significant detection
in the V band. We argue however that foregrounds and in particular point sources are the origin of
this signal.

I. INTRODUCTION

The Wilkinson Microwave Anisotropy Probe (WMAP) data [1, 2] have raised a number of interrogations concern-
ing the statistical properties of the temperature field. While these data globally confirm the standard inflationary
paradigm [3] and the concordance cosmological model, they exhibit some intriguing anomalies, particularly concerning
the large angular scales. In particular, a huge activity has been devoted to the study of the low value of the quadrupole
and octopole [4, 5, 6, 7] as well as their alignment [8, 9], two effects that appear to be inconsistent with the standard
cosmological model.
Besides, many authors have tried to test the statistical properties of the temperature field using various methods.
For instance, it was investigated whether the coefficients aℓm of the development of the temperature field on spherical
harmonics were independent and Gaussian distributed. While, as expected from the standard inflationary picture, a
χ2 deviation from Gaussianity seems to be well constrained [10], there have been some claims that the distribution
may not be isotropic [9, 11, 12, 13, 14, 15] or Gaussian [16, 17, 18, 19]. No real physical understanding of these
measurements have been proposed yet and the origin of these possible features is still unknown. Some authors have
argued in favor of systematic effects [13] while it was argue [14, 20] that foreground contamination may play an
important role in these conclusions.
From a theoretical point of view, there are many reasons to look for (and/or constrain) a departure from Gaussianity
and/or isotropy of the CMB temperature field. Mode correlation can be linked to non-Gaussianity, in particular due
to finite size effects [21, 22, 23] or to the existence of some non-trivial topology of the universe [25]. While in the
latter case, one expects to have a complex correlation matrix of the aℓm , the former leads generically to a dipolar
modulation of the CMB field [24]. Such a modulation induces in particular correlations between adjacent multipoles
(∆ℓ = 1). Similar correlations but with ∆ℓ = 2 may also be induced by a primordial magnetic field [26]. In each
case, the physical model and its predictions indicate the type of correlations to look for and will drive the design of
an adapted estimator.

∗ Electronic address: prunet@iap.fr

† Electronic address: uzan@th.u-psud.fr,uzan@iap.fr
‡ Electronic address: fbernard@spht.saclay.cea.fr
§ Electronic address: brunier@spht.saclay.cea.fr
 2

Investigation of the correlation properties of aℓm is thus important to correctly interpret previous observational
results [9, 11, 12, 13, 14, 15]. Two approaches are thus possible. Either one defines some general estimators and study
whether they agree with a Gaussian and isotropic distribution (top-down approach) or one sticks to a class of physical
models and construct an adapted estimator (bottom-up approach). In this article, we follow the second route and
focus to the task of constraining a possible dipolar modulations of the CMB temperature field, that is correlations
between multipoles with ∆ℓ = 1.
In Section II, we start by some general considerations on the form of the correlation arising from a dipolar mod-
ulation. We then built an estimator, in Section III, adapted to these types of correlations. In particular, we cannot
use full-sky data and we will need to cut out some part of the sky. The effect of such a mask on the correlations will
have to be taken into account and included in the construction of the estimator. We apply this estimator to the V
and W bands of the WMAP data in Section IV. The V band exhibits an apparent detection. The interpretation of
this result will require us to compare various masks, and in particular to investigate the effect of point sources on the
signal to conclude that they are most likely its cause.

II. GENERAL CONSIDERATIONS

As explained in the introduction, we focus on a possible dipolar modulation of the CMB signal. Thus, we assume
that the observed temperature field can be modelled as

Θobs (~γ ) = Θ(~γ ) [1 + ε−1 Y1,−1 (~γ ) + ε0 Y1,0 (~γ ) + ε+1 Y1,+1 (~γ )] (1)

where Θ is the genuine statistically isotropic field and where (ε−1 , ε0 , ε+1 ) are three unknown parameters that char-
acterizes the direction of the modulation. The modulation has to be real so that ε0 is real and ε+1 = −ε−1 ∗ ≡ ε.
As usual, we decompose the temperature fluctuation in spherical harmonics as
∞ X
X ℓ
Θ(~γ ) = aℓm Yℓ,m (~γ ). (2)
ℓ=2 m=−ℓ

The coefficients aℓm are thus given by

Z
aℓm = d2~γ Θ(~γ )Yℓ,m
∗
(~γ ). (3)

Θobs and aobsℓm are defined and related in the same way. Since Θ is supposed to be the primordial, Gaussian and
statistically isotropic, temperature field, its correlation matrix reduces to

haℓm a∗ℓ′ m′ i = Cℓ δℓℓ′ δmm′ . (4)

A modulation of the form (1) implies that the coefficients aobs

ℓm develop correlations between multipoles with ∆ℓ = 1.
Let us illustrate the origin of this correlation. From Eqs. (1) and (4), we deduce that

∞
X ℓ
X
′
+1
X Z
aobs
ℓm = aℓm + aℓ ′ m′ εi d2~γ Yℓ,m
∗
(~γ )Yℓ′ ,m′ (~γ )Y1,i (~γ ). (5)
ℓ′ =2 m′ =−ℓ′ i=−1

The integral can be easily computed by using the Gaunt formula [see Eq. (A7)] to get
r   
obs 3 X m
X p ℓ L 1 ℓ L 1
aℓm = aℓm + εi (−1) aLM (2ℓ + 1)(2L + 1) . (6)
4π i −m M i 0 0 0
LM

Because of the triangular inequality, the Wigner 3j-symbols are non zero only when L = ℓ ± 1 and M = m − i so
that aobs
ℓm is in fact a sum involving aℓm and aℓ±1m−i . It follows that it will develop ℓ − (ℓ + 1) correlations that can
be characterized by the two quantities
(0)
obs obs∗ 
Dℓm ≡ aℓm aℓ+1m , (7)
(1)
obs obs∗ 
Dℓm ≡ aℓm aℓ+1m+1 (8)
 3

which will be non zero respectively as soon as ε0 or ε are non zero. Using the expression (6) and the property (4) of
the primordial field, we deduce that
r p
(0) 3 (ℓ + 1)2 − m2
Dℓm = ε0 p (Cℓ + Cℓ+1 ) (9)
4π (2ℓ + 1)(2ℓ + 3)

r s
(1) 3 (ℓ + 2 + m)(ℓ + 1 + m) ε∗
Dℓm = [Cℓ + Cℓ+1 ] √ . (10)
4π (2ℓ + 1)(2ℓ + 3) 2

Interestingly, these forms indicate how to sum the Dℓm in order to construct an estimator. This construction will be
detailed in the following section.

III. MATHEMATICAL CONSTRUCTION OF THE ESTIMATOR

The previous analysis is illustrative but not suitable to be applied on real data. In particular these data will not
be full sky and we have to take into account the effect of a mask (see e.g. Ref. [27]). Such a mask, that arises in
particular because of the galactic cut, will induce correlations in the coefficients aobs
ℓm that are described in § III A.
We design the mask in order to protect the correlations that originate from the modulation (§ III B and § III C) and
finish by presenting the construction of our estimator in the most general case (§ III D).

A. Mask effects

The temperature field is observed only on a fraction of the sky. We thus have to mask part of the map so that the
temperature field is in fact given by
" 1
#
X
Θobs (~γ ) = Θ(~γ ) 1 + εi Y1,i (~γ ) W (~γ ) (11)
i=−1

where W (~γ ) is a window function, referred to as mask, indicating which part of the sky has been cut. We decompose
W (~γ ) in spherical harmonics as
X
W (~γ ) = wℓm Yℓm (~γ ). (12)
ℓm

∗
W (~γ ) being a real valued function, it implies that wℓm = (−1)m wℓ−m . We deduce from Eqs. (11) and (3) that
X (i)
aobs
ℓm = eaℓm + εi Aℓm (13)
i

where e e γ ) = Θ(~γ )W (~γ ),

aℓm are the coefficients of the masked primordial temperature field Θ(~
X X Z
aℓm =
e aℓ 1 m1 wℓ2 m2 d2~γ Yℓ1 m1 (~γ )Yℓ2 m2 (~γ )Yℓ,m
∗
(~γ ) (14)
ℓ 1 m1 ℓ 2 m2

and the effects of the modulation are encoded in the correction

X X Z
(i)
Aℓm = aℓ 1 m1 wℓ2 m2 d2~γ Yℓ1 m1 (~γ )Yℓ2 m2 (~γ )Y1i (~γ )Yℓ,m
∗
(~γ ). (15)
ℓ 1 m1 ℓ 2 m2

ℓ 1 m1
Interestingly, e
aℓm can be shown to be obtained from aℓm by the action of a Kernel Kℓm
X
ℓ 1 m1
aℓm =
e aℓ1 m1 Kℓm . (16)
ℓ 1 m1
 4

This kernel is defined by

X Z
ℓ 1 m1
Kℓm ≡ wℓ2 m2 d2~γ Yℓ1 m1 (~γ )Yℓ2 m2 (~γ )Yℓ,m
∗
(~γ )
ℓ 2 m2

and can be explicitly computed by using the integral (A7) to obtain

r   
X (2ℓ1 + 1)(2ℓ2 + 1)(2ℓ + 1) ℓ1 ℓ2 ℓ ℓ1 ℓ2 ℓ
ℓ 1 m1 m
Kℓm = (−1) wℓ2 m2 . (17)
4π m1 m2 −m 0 0 0
ℓ 2 m2

The contribution arising from the modulation can be computed by using the integral (A7) to get
r   
(i) 3 X √ √ L 1 ℓ L 1 ℓ
Aℓm = (−1)m eaLM 2ℓ + 1 2L + 1 . (18)
4π M i −m 0 0 0
LM

One can check that the relation (5) obtained without taking into account the effects of the mask still holds if one
replaces aℓm by eaℓm . The complications arise from the fact that e
aℓm does not satisfy the property (4) because of the
action (16) of the Kernel.

B. Choice of the mask and properties of the masked quantities

We now need to specify the form of the mask. First, let us note that when W (~γ ) =constant for all ~γ then one
trivially recovers that e
aℓm = aℓm because W = w00 Y00 so that
ℓ 1 m1 w00
Kℓm = √ δℓℓ1 δmm1 .
4π
Since we are looking for ℓ − (ℓ + 1) correlations, we would like to design a mask that does not involve the same
correlations for e
aℓm and that is not m-dependent. A solution is to impose that W (~γ ) is a function of θ only and that
it is north-south symmetric, that is
W (~γ ) = W (θ), W (π − θ) = W (θ). (19)
Since Yℓ0 (π − θ) = (−1)ℓ Yℓ0 (θ), these conditions imply that
X [1 + (−1)ℓ2 ] X
W (~γ ) = wℓ2 Yℓ2 0 (θ) ≡ ŵℓ2 Yℓ2 0 (θ) (20)
2
ℓ2 ℓ2

The simplest example of such a mask is obtained by considering a function which is constant and vanishes on an
equatorial strip of latitude θc ∈ [0, π/2]. This implies that the multipoles of the mask are given by
√
w0 = 4πµc , (21)
r ℓ
4π [1 + (−1) ]
ŵℓ = [Pℓ−1 (µc ) − Pℓ+1 (µc )] (22)
2ℓ + 1 2
where µc = cos θc . In
√ particular, it can be seen that when θc → 0, that is when the size of the mask vanishes, this
mask satisfies wℓ → 4πδℓ0 when µc → 1. The function wℓ is depicted on figure 1 for galactic cuts of 10, 20 and 30
degrees.
The results derived in the following sections are not dependent on the particular choice of the mask as long as it
satisfies the symmetries (19) which ensure that the coefficients of the mask do not depend on m and vanish for ℓ odd
(see Eq. 20).

C. Properties of the e
aℓm

Whatever the choice of the mask, as long as it satisfies the properties (19), the general expression of the coefficients
e are given by
aℓm of the decomposition of Θ,
e
r   
√ X 2ℓ1 + 1 Xp ℓ1 ℓ2 ℓ ℓ1 ℓ2 ℓ
m
aℓm = fsky aℓm + (−1)
e 2ℓ + 1 aℓ 1 m 2ℓ2 + 1ŵℓ2 (23)
4π m 0 −m 0 0 0
ℓ1 ℓ2 6=0
 5

0.6

0.4

0.2

-0.2

-0.4
0 5 10 15 20

FIG. 1: (left) The coefficients of the decomposition of the mask function on spherical harmonics for different galactic cuts,
µc = 10, 20 and 30 deg. (respectively black, red and blue lines). Only the non-vanishing, that is even, multipoles are plotted.
(right) The ratio wℓ /w0 decreases rapidly and becomes typically smaller than 0.1% for ℓ larger than 20 (plain=10 deg., dot=20
deg., dash-dot=30 deg.).

√
with fsky ≡ w0 / 4π is the fraction of the sky that is covered. From this expression, we deduce that their 2-point
function is given by
( )
X
∗ 2 ′ ′ ′
he
aℓm e
aℓ′ m′ i = δmm′ Cℓ fsky δℓℓ′ + fsky [G(ℓ, ℓ , m)Cℓ + G(ℓ , ℓ, m)Cℓ′ ] + Cℓ1 G(ℓ1 , ℓ, m)G(ℓ1 , ℓ , m) (24)
ℓ1

where the function G(ℓ, ℓ′ , m) is defined by

r   
p X 2ℓ2 + 1 ℓ ℓ2 ℓ′ ℓ ℓ2 ℓ′
′ m
G(ℓ, ℓ , m) = (−1) (2ℓ + 1)(2ℓ′ + 1) ŵℓ2 . (25)
4π m 0 −m 0 0 0
ℓ2 6=0

It follows from Eq. (24) that there is no m-coupling arising from the mask (because it has no azimuthal dependence)
and we can define the correlation matrix of the masked temperature field as
he
aℓm e eℓm δmm′ .
a∗ℓm′ i ≡ C (26)
The angular power spectrum of the mask field is then defined as
ℓ
X
eℓ = 1 eℓm
C C (27)
2ℓ + 1
m=−ℓ

and is explicitly given in terms of the primordial angular power spectrum by

" ℓ
# ℓ
f sky
X X Cℓ X
e 2
Cℓ = Cℓ fsky + 2 G(ℓ, ℓ, m) + 1
G 2 (ℓ1 , ℓ, m). (28)
2ℓ + 1 2ℓ + 1
m=−ℓ ℓ1 m=−ℓ

Let us now turn to the ℓ-(ℓ + 1) correlators. The first term in Eq. (24) vanishes. Then, one can check that
G(ℓ, ℓ + 1, m) vanishes because the triangular relation of the Wigner-3j symbols implies that ℓ2 = ±1 but for odd ℓ2
ŵℓ2 vanish. To finish, the contribution of G(ℓ1 , ℓ, m)G(ℓ1 , ℓ + 1, m) in the sum also vanishes because ℓ2 is even and
the sums ℓ1 + ℓ2 + ℓ + 1 and ℓ1 + ℓ2 + ℓ have to be both even, which is impossible. In conclusion


e a∗ℓ+1m′ = 0.
aℓm e (29)
As expected from our construction, the mask does not generate ℓ-(ℓ + 1) correlations.
To finish, let us stress that the mask will induce some ℓ-ℓ + 2 correlations that can be characterized by introducing
ℓ
X

 1
e
2 Cℓm ≡ e a∗ℓ+2m ,
aℓm e e
2 Cℓ ≡
e
2 Cℓm . (30)
2ℓ + 1
m=−ℓ

eℓ = 0.
Indeed, when W = Id, 2 C
 6

D. General construction

Starting from the relation (13) and the expression (18), we deduce that the two quantities defined in Eqs. (7-8)
generalize to
r   h i
(i) 3 ∗ √ ℓ 1 ℓ+1
Dℓm = εi (−1) ℓ+m+1+i
ℓ+1 eℓm + C
C eℓ+1m+i
4π m i −m − i
 
√ ℓ−1 1 ℓ
+(−1) ℓ+m+i
ℓ Ce
m + i −i −m 2 ℓ−1m+i
  
√ ℓ+1 1 ℓ+2
+(−1) ℓ+m+i
ℓ+2 e
2 Cℓm , (31)
−m − i i m

with i = 0, 1, when the mask effects are taken into account. This expression is defined for m = −ℓ . . . ℓ even if
e e
2 Cℓ−1m+1 is not defined for m = ℓ and m = ℓ − 1 and 2 Cℓ−1m for m = ℓ because the Wigner-3j symbols that
multiply these terms strictly vanish. From this expression, we define
 
Pℓ ℓ+m+1+i ℓ 1 ℓ+1 (i)
m=−ℓ (−1) Dℓm
(i) m i −m − i
Dℓ ≡   . (32)
Pℓ ℓ+m+1+i ℓ 1 ℓ+1
m=−ℓ (−1)
m i −m − i

Now, it can be checked, after some algebra, that

r
(i) 3 ∗ h (i) (i)A (i)B
i
Dℓ = εi Ĉℓ + 2 2 Ĉℓ + 2 2 Ĉℓ + O(ε2 ) (33)
4π
(i) (i)A (i)B
where the quantities Ĉℓ , 2 Ĉℓ and 2 Ĉℓ have been defined by
ℓ
X  2 h i
(i) 1 p ℓ+1 1 ℓ
Ĉℓ = ℓ + 1) Ceℓm + C
eℓ+1m+i (34)
(i)
Nℓ m i −m − i
m=−ℓ
ℓ
X   
(i)A 1 √ ℓ 1 ℓ+1 ℓ+2 1 ℓ+1
2 Ĉℓ = − ℓ+2 e
2 Cℓm (35)
Nℓ
(i) m i −m − i m i −m − i
m=−ℓ
ℓ
X   
(i)B 1 √ ℓ 1 ℓ+1 ℓ−1 1 ℓ
2 Ĉℓ = − ℓ e
C (36)
Nℓ
(i) m i −m − i m + i −i −m 2 ℓm
m=−ℓ

(i)
with the coefficients Nℓ given by
ℓ
X  
(i) ℓ 1 ℓ+1
Nℓ = (−1)ℓ+m+i . (37)
m i −m − i
m=−ℓ

It follows from these results that we can consider the estimator

ℓ
X  
(i) 1 ℓ 1 ℓ+1
Eℓ = (−1)ℓ+m+i aobs obs∗
ℓm aℓ+1m+i (38)
Nℓ
(i) m i −m − i
m=−ℓ

that satisfies by construction

D E
(i) (i)
Eℓ = Dℓ . (39)

We will apply this estimator to the WMAP data in the following sections.
 7

FIG. 2: Dℓ measured on the WMAP data (W band) [green triangles]. Blue triangles are the measurements on a simulated map
with ε0 = 0.2 and the red lines are 1σ error bars.

IV. DATA ANALYSIS

The proposed estimators have been implemented numerically, using the Healpix [32] package for the pixelization
and the fast spherical harmonics transforms, and applied to the co-added data of the WMAP V and W bands (resp.
70 GHz and 90 GHz) where most of the signal is of cosmological origin. We implemented the estimators as described
by equations (33) to (38).
(i) (i)A (i)B
The quantities Ĉℓ , 2 Ĉℓ and 2 Ĉℓ have been computed using the best fit LCDM theoretical power spectrum
of the WMAP data [28], and were not computed on the data itself to avoid ratios of random variables. To assess the
statistical significance of the measured values of εi , we made 1000 simulations of WMAP data in each of the V and
W bands according to a sky model with no modulation.
The results of the analysis of the V and W bands are summarized on Figs. 2 to 5. Fig. 2 depicts the measurement
of Dℓ on the W band. We sum this measurement of two bands of ℓ (respectively 20-100 and 100-300) and compare
with 1000 simulated WMAP data. We perform the same tests on the V band (Fig. 4 and 5). The apparent detection
in the V band without clear counterpart in the W band suggest a non cosmological contamination. Determining its
origin requires to performed more tests.
This contamination can be a priori from two possible sources, Galactic or extragalactic. To check if the correlations
detected in the V band are of Galactic origin, we apply the same estimator to the half sum and half difference of the
V and W bands, that is
W +V W −V
S= , D= (40)
2 2
and repeat the whole procedure on 1000 simulation in each case, where the simulations contain only CMB and noise
according to the WMAP specifications. The advantage of the half difference of the bands is that it should (up to
calibration errors) eliminate the CMB signal completely at large scales, hence eliminate the main source of variance
at these same scales, where the Galactic signals are expected to dominate. Indeed, the power spectra of Galactic
emissions usually scale as Cℓ ∝ ℓ−α , with 2 ≤ α ≤ 3 (see e.g. Ref. [29]). The half sum results, summarized in table I,
are in between those of the V and W bands, which is coherent with the assumption of the detection being caused by
a foreground source of electromagnetic spectrum different from the CMB fluctuations.
More importantly, the half difference results do not show a strong correlation detection at large angular scales,
in contradiction with the assumption of the Galactic foreground contamination being the source of the detected
correlations in the V band.
However, this half difference test does not work that well if the source contaminants are of extragalactic origin,
since the power spectra of extragalactic foregrounds resemble that of the noise. In this case, the contamination is
expected to increase with increasing multipole number, which seems to be the case for the V band (see Figs. 4 and
5).
The difficulty of extragalactic point sources contamination is that these sources (quasars and active radio-galaxies)
are distributed more or less uniformly across the sky, which renders their masking by an azimuthally symmetric sky
cut impossible. However, the WMAP team provides with their data sets “taylor cuts” that blank out the resolved
 8

FIG. 3: Comparison of 1000 simulations with the WMAP data. We use the W band and sum the multipole between ℓ = 20
and ℓ = 100 (top panel) and between ℓ = 100 and ℓ = 300 (bottom panel). The color code is identical to the one of Fig. 2 that
is the green line correspond to the measurement on the WMAP data and the blue line the measurement on a simulation.

point sources of largest flux. Of course, the dipolar modulation estimators designed in the preceding sections do
not apply stricto sensu to these arbitrary masks, but one can hope, given the small fraction of sky removed at high
latitude, that the broken symmetry of the mask will be a small perturbation in the computation of the ε’s, so that
the estimators keep their general validity, up to a possible small bias (see Fig. 8 for a comparison of the coefficients
wℓ of the two masks).
This assumption can be checked on a simulated sky with a known dipolar modulation, where a WMAP V-band
noise is added to the signal, together with the taylor mask. We chose the most conservative mask provided by the
WMAP team (kp0), and applied it to a simulated sky of known modulation (ε0 = 0.2) as described above, together
with the V-band data. The results are shown in figures 6 and 7. Again, the estimators have been applied to 1000
simulations of the V band with no dipolar modulation, with the same kp0 mask applied, to estimate the statistics of
the V-band data results.
Several observations can be made at this point. First, comparing these results with those obtained in the V-band,
but with the azimuthally symmetric 20◦ cut (figure 4), one can check that the estimators give very compatible results
for the simulated dipolar modulation (blue squares). This comforts our assumption that changing the cut sky to the
kp0 mask is a small perturbation for the modulation estimators.
Secondly, comparing the same figures but this time looking at the data (green squares), one can see that in the case
of the 20◦ cut there is a large trend at high ℓ’s in ε1 that disappears when the kp0 cut is used. This is confirmed by
the results of table I where one can check that the tentative detections of in the V band using the simple cut become
statistically insignificant when using the kp0 cut.
 9

V. DISCUSSION AND CONCLUSIONS

In this article we have proposed an estimator designed to detect a possible modulation of the CMB temperature
field, or equivalently ∆ℓ = 1 correlations. The effects of cutting part of the sky were discussed in details and we
applied this estimator to the V and W bands of the WMAP data.
The results of our analysis are summarized in table I which gives the amplitude of the modulation coefficients on
the WMAP data and a corresponding test case with ε0 = 0.2, Re(ε1 ) = Im(ε1 ) = 0. All values are given in standard
deviation units, estimated on 1000 (signal+noise) simulations in each case, with no modulations.
While the V band seems to exhibit a marginal detection, further tests such as the study of the half sum and
difference of the two bands and the effect of point sources have led us to conclude that this detection should be
inferred to the effect of point sources contamination. In this analysis we have used the kp0 mask which does not
satisfied the symmetries of the mask required for our estimator to be unbiased. Nevertheless, our estimator seems to
be well suited for the analysis, even with the kp0 mask.
To back up this interpretation we have performed two last tests. First we added to a simulated CMB map without
modulation and with noise the 208 sources resolved by the WMAP experiment and then smoothed with the correct
beam. Second we added to the same simulation the 700 circular region that are cut in the analysis of the V band
in the WMAP analysis. Both simulations, while analyzed as the previous data with an azimuthal mask of 20 deg.,
exhibit an excess of signal for ε0 and Im(ε1 ) in the same range of multipoles than obtained on the analysis of the V
and W bands (Figs. 2 and 4). Interestingly, the signal of Re(ε1 ) is not affected and is identical to the one of Figs. 2
and 4. Indeed, the signals have not exactly the same amplitude as the ones obtained from the analysis of the V band
but they exhibit the same trend on the the same scales. Also, it has to be stressed that with a cut of 20 deg. the Large
Magellannic Cloud (galactic latitude of 20 deg. and more and longitude of 0 deg.) and a part of the H2 Ophucius
region should contribute and that we have not included them in the simulations. This could have enhance the signal.
In conclusion, the set of analysis performed in our study tend to show that the ∆ℓ = 1 correlations that appeared
in the analysis of the V and W bands of the WMAP data are due to foreground contaminations and most likely by
point sources. The direction of the detected modulation will, in that interpretation, characterize the anisotropy of
the distribution of these sources.

Acknowledgements: Some of the results of this article have been derived using the HEALPix package [30]. We
thank Y. Mellier and R. Stompor for discussions.

APPENDIX A: INTEGRALS OVER SPHERICAL HARMONICS

We have evaluated integrals over n spherical harmonics (see Ref. [31]). When n = 1 or 2, these integrals are trivial
Z √
d2 ~γ Yℓm = 4πδℓ0 δm0 (A1)

FIG. 4: Dℓ measured on the WMAP data (V band) [green traingles]. Blue triangles are the measurements on a simulated map
with ε0 = 0.2 and the red lines are 1σ error bars.
 10
Z
d2 ~γ Yℓ1 m1 Yℓ∗2 m2 = δℓ1 ℓ2 δm1 m2 . (A2)

To go further, one solution is to use the decomposition of the product of 2 spherical harmonics as
s
X (2ℓ1 + 1)(2ℓ2 + 1
Yℓ1 m1 (~γ )Yℓ2 m2 (~γ ) = CℓL0
1 0ℓ2 0
CℓLM Y (~γ )
1 m1 ℓ2 m2 LM
(A3)
4π(2L + 1)
LM

where the CℓLM

1 m1 ℓ 2 m2
are the Clebsch-Gordan coefficients that can be expressed in terms of Wigner 3j symbols as
!
LM ℓ1 −ℓ2 +M
√ ℓ1 ℓ2 L
Cℓ1 m1 ℓ2 m2 = (−1) 2L + 1 (A4)
m1 m2 −M

It is easy to generalize Eq. (A3) to a product of n spherical harmonics

 r !
r n
X 4π X Y 2ℓ i + 1 i0
Yℓ1 m1 . . . Yℓn mn =  CLLi−1 Li Mi
0ℓi 0 CLi−1 Mi−1 ℓi mi
 YLn Mn . (A5)
2Ln + 1 i=1
4π
Ln ,Mn L1 ...Ln−1 ,M1 ...Mn−1

We deduce, using Eq. (A5) and the integral (A2) that

Z s
(2ℓ1 + 1)(2ℓ2 + 1) ℓ3 0
2 ∗
d ~γ Yℓ1 m1 Yℓ2 m2 Yℓ3 m3 = Cℓ1 0ℓ2 0 Cℓℓ13m
m3
1 ℓ 2 m2
(A6)
4π(2ℓ3 + 1)

FIG. 5: Comparison of 1000 simulations with the WMAP data. We use the V band and sum the multipole between ℓ = 20
and ℓ = 100 (top panel) and between ℓ = 100 and ℓ = 300 (bottom panel).
 11

FIG. 6: Dℓ measured on the WMAP data (V band) [green triangles], using the taylor mask kp0 to blank the main point
sources. Blue triangles are the measurements on a simulated map with ε0 = 0.2 and the red lines are 1σ error bars.

ε0 Re(ε) Im(ε)
data test data test data test
W (20-100) -0.45 5.87 1.30 -0.26 2.54 -1.14
W (100-300) -0.60 16.9 1.65 0.61 0.59 0.41
V (20-100) -0.04 6.00 1.61 -0.17 3.21 -1.03
V (100-300) -0.81 17.9 0.25 0.65 4.10 0.54
V-kP0 20-100) -0.11 6.12 1.83 0.16 2.20 -0.42
V-kp0 (100-300) -0.89 17.4 1.98 1.45 -0.22 0.18
S (20-100) -0.24 6.71 1.52 0.40 2.85 -0.31
S (100-300) -0.64 19.3 1.15 0.57 2.16 1.35
D (20-100) -0.58 -0.74 -2.10 -1.49 3.73 -0.70
D (100-300) -0.98 0.93 -0.44 0.69 2.67 -0.54

TABLE I: Summary of the data analysis performed in this article. It concerns the two bands V and W, their half sum (S) and
half difference (D). The band V has been analyzed with two masks to emphasize the effect of the point sources on the result.

Z s
X (2ℓ1 + 1)(2ℓ2 + 1)(2ℓ3 + 1) L0
2 ℓ4 0 ℓ 4 m4
d ~γ Yℓ1 m1 Yℓ2 m2 Yℓ3 m3 Yℓ∗4 m4 = Cℓ1 0ℓ2 0 CL0ℓ 30
CℓLM
1 m1 ℓ 2 m2
CLMℓ 3 m3
. (A7)
(4π)2 (2ℓ4 + 1)
L,M

[1] C.L. Bennett et al., Astrophys. J. Suppl. 148 (2003) 1.

[2] D.N. Spergel et al., Astrophys. J. Suppl. 148 (2003) 175.
[3] H.V. Peiris et al., Astrophys. J. Suppl. 148 (2003) 213.
[4] M. Tegmark, A. de Oliveira-Costa, and A. Hamilton, Phys. Rev. D 68 (2003) 123503.
[5] J.-P. Uzan, U. Kirchner, and G.F.R. Ellis, Month. Not. R. Astron. Soc. 343 (2003) L95.
[6] G. Efstathiou, Month. Not. R. Astron. Soc. 343 (2003) L95.
[7] A. Slosar, U. Seljak, and A. Makarov, [arXiv:astro-ph/0403073].
[8] A. de Oliveira-Costa, M. Tegmark, M. Zaldarriaga, and A. Hamilton, Phys. Rev. D 69 (3004) 63516.
[9] D. Schwarz et al., [arXiv:astro-ph/0403353].
[10] E. Komatsu et al., Astrophys. J. Suppl. 148 (2003) 119.
[11] A. Haijan and T. Souradeep, Astrophys. J. 597 (2003) L5.
[12] C.J. Copi, D. Huterer, and G. Starkman, [arXiv:astro-ph/0310511].
[13] F.K. Hansen, P. Cabella, D. Marinucci, and N. Vittorio, [arXiv:astro-ph/0402396].
[14] H.K. Eriksen, D.I. Novikov, P.B. Lilje, A.J. Banday, and K.M. Gorski, [arXiv:astro-ph/0401276].
[15] H.K. Eriksen, F.K. Hansen, A.J. Banday, K.M. Gorski, and P.B. Lilje, [arXiv:astro-ph/0307507].
[16] C.-G. Park, Month. Not. R. Astron. Soc. 349 (2004) 313.
 12

FIG. 7: Comparison of 1000 simulations with the WMAP data. We use the V band and sum the multipole between ℓ = 20
and ℓ = 100 (top panel) and between ℓ = 100 and ℓ = 300 (bottom panel) using the taylor mask kp0 to blank the main point
sources.

[17] D.L. Larson and B. Wandelt, [arXiv:astro-ph/0404037].

[18] L.-Y. Chiang, P.D. Naselsky, O. Verkhodanov, and M. Way, Astrophys. J. Lett 590 (2003) 65.
[19] K. Land and J. Magueijo, [arXiv:astro-ph/0405519].
[20] A. Slosar and U. Seljak, [arXiv:astro-ph/0404567].
[21] F. Bernardeau and J.-P. Uzan, Phys. Rev. D 66 (2002) 103506.
[22] F. Bernardeau and J.-P. Uzan, Phys. Rev. D 67 (2003) 121301(R).
[23] F. Bernardeau and J.-P. Uzan, Phys. Rev. D (in press), [arXiv:astro-ph/0311421].
[24] F. Bernardeau, T. Brunier and J.-P. Uzan, (in preparation).
[25] A. Riazuelo, J.-P. Uzan, R. Lehoucq, J. Weeks, Phys. Rev. D 69 (2004) 103514;
A. Riazuelo, J.-P. Uzan, R. Lehoucq, J. Weeks, and J.-P. Luminet, Phys. Rev. D 69 (2004) 103518;
J.-P. Uzan and A. Riazuelo, C.R. Acad. Sciences (Paris) 4 (2003) 945;
J.-P. Uzan, A. Riazuelo, R. Lehoucq, and J. Weeks, Phys. Rev. D 69 (2004) 043003,
J.-P. Luminet, J. Weeks, A. Riazuelo, R. Lehoucq, and J.-P. Uzan, Nature (London) 425 (2003) 593.
[26] R. Durrer, T. Kahniashvili, and T.A. Yates, Phys. Rev D 58 (1998) 123004;
A. Mack, T. Kahniashvili, and A. Kosowsky, Phys. Rev. D 65 (2002) 123004;
G. Chen, P. Mukherjee, T. Kahniashvili, B. Ratra, and Y. Wang, [arXiv:astro-ph/0403695];
P.D. Naselsky, L.-Y. Chiang, P. Olesen, and O. Verkhodanov, [arXiv:astro-ph/0405181].
[27] E. Hivon et al., Astrophys. J. 567 (2002) 2.
[28] L. Verde et al., Astrophys. J. Suppl. 148 (2003) 195.
[29] F.R. Bouchet and Gispert, New Astron. 4 (1999) 443.
[30] K.M. Górski, E. Hivon, and B.D. Wandelt, in Proceedings of the MPA/ESO Cosmology Conference ”Evolution of large-
scale structure”, Eds. A.J. Banday et al. (PrintPartners Ipskamp, NL, 1999), pp. 37-42, [arXiv:astro-ph/9812350].
[31] D.A. Varshalovich, A.N. Moskalev, and V.K. Khersonskii, “Quantum theory of angular momentum”, (World Scientific,
Singapore, 1988).
[32] http://www.eso.org/science/healpix
 13

FIG. 8: The coefficients of of the decomposition of the mask function on spherical harmonics for different galactic µc = 10, 20
and 30 deg. (respectively black, red and blue lines) compared with the ones of the kp0 mask (green).
 14

FIG. 9: The analysis of the simulated maps in which (top) the 208 resolved sources of the WMAP catalog have been added
and (bottom) where the 700 point sources (resolved and unresolved) of the V band have been added. In both cases, the signal
of Re(ε1 is not affected while the signal of ε0 and Im(ε1 ) exhibit patterns that are similar to the ones obtained in our analysis
of the V and W band on the same angular scales.